Imaging mouse whisking using an event camera

One of the greatest challenges in rodent neuroscience is accurately monitoring whisking behavior in freely moving animals within naturalistic environments. To understand the neural computations underlying perception and behavior, neuroscientists need precise temporal registration of an animal’s interactions—such as whisker touches—with simultaneous neural recordings. However, achieving this is technically difficult. Rodents move rapidly and unpredictably, often positioning their heads at variable angles during exploration. These behaviors occur in low-light environments and are highly idiosyncratic across individuals, adding further complexity. Compounding this, whisking is a fast sensorimotor behavior, often occurring at frequencies at approximately 20Hz, which requires high-frame-rate video recordings to resolve individual whisker-object contacts. These high frame rates, in turn, generate extremely large datasets, creating logistical and computational bottlenecks for data storage, processing, and analysis. All of these factors make precise, temporally aligned behavioral and neural monitoring a significant technical hurdle. Here we have optimized a way to address many of these technical hurdles by using an event-based camera. By only capturing pixels that have changes in illumination, we can massively reduce the data rate. Already completed work on this research has demonstrated that there is a unique temporal signature for mouse whisking and that this signal is strong relative to the noise floor of current performance event-based imagers. Discussions on lighting, shadows, lens focus, and whisker reflection will support a needed infrastructure for performing this task using an event-based imager. Additionally, there are AI/ML algorithms for tracking multiple mice in an open environment using traditional frame-based imagers. These are accepted tools for tracking mouse movement, but this research intends to augment these algorithms in two unique ways. First, these AI/ML algorithms are trained on and expecting either monochromatic or RGB color imagery. Some work may be necessary for these algorithms to ingest event data instead. Secondly, if a mouse is tracked and oriented using event-based imagery, regions of interest can be placed around a mouse’s whiskers for high frequency pose and motion estimation of whiskers.